Fire-fighting spray header and detection and early warning system based on fire-fighting spray header

Abstract

The invention provides a fire-fighting spray head and a detection early warning system based on the fire-fighting spray head. An inner pipe is arranged in a fire-fighting pipeline, supporting is provided through a supporting rod, a sealing disc and a fire-fighting glass ball in sequence, and the inner pipe is arranged at the position with the proper height away from the end of the fire-fighting pipeline through a connecting section head. The connecting arm attached to the periphery of the fire-fighting pipeline is arranged through the sealing disc, the position of the detection through hole of the inner pipe is marked on the fire-fighting pipeline through the connecting arm, and the external sensing detection device is installed. According to the fire-fighting spray head, on one hand, the detection through holes of the inner pipe can be used for carrying out a flue gas detection test to determine the mounting effect of each spray head, and on the other hand, the corresponding sensing detection device can be continuously used for dynamically monitoring each mounting point position in a building after the flue gas detection test is finished; and the fire risk of each point location in the building area is obtained by combining the transmission characteristics of the fire in the building, so that early warning is effectively realized.

Description

Technical Field

[0001] The present application relates to the technical field of fire detection equipment, and in particular to a fire sprinkler head and a detection and early warning system based thereon. Background Art

[0002] The existing fire sprinkler system has a single function. The existing fire sprinkler head has an organic solution with a high thermal expansion coefficient in the glass ball of the glass ball sprinkler. At room temperature, the outer shell of the ball can withstand a certain support force to ensure the sealing performance of the sprinkler head; when a fire occurs or an external force damages it, the temperature of the organic solution in the glass ball rises and expands until the glass body breaks, and the ball seat and the seal lose their support and are washed away by the water flow, thus starting to spray water to extinguish the fire.

[0003] The existing fire glass ball, its bursting and spraying action is only triggered by temperature conditions, and cannot respond to other environmental characteristics. In addition, the existing fire sprinkler device is a one-time device, has no other functions before spraying, and becomes invalid after spraying, and cannot achieve any other functions.

[0004] In addition, due to the failure characteristics of fire sprinkler heads, reliability testing through water spray drills after installation will greatly affect the use environment of the work and living places where they are located, causing damage to objects, documents, circuits, electronic equipment, etc. Other manual inspection methods require the removal of the sprinkler pipe sealing layer (such as the ceiling), and the inspection process is complicated and greatly affected by environmental factors, and many areas will be difficult to detect. Summary of the invention

[0005] In view of the deficiencies of the prior art, the present application provides a fire sprinkler head and a detection and early warning system based thereon. The present application integrates a sensor detection device on the fire sprinkler head, and transmits the sensor data detected by the sensor detection device in real time to a pre-trained detection model. The detection model obtains the fire risk of each point in the building area according to the installation point of each sensor detection device, the sensor data, and the internal characteristics of the building obtained through the previous effectiveness test to achieve early warning. The present application specifically adopts the following technical solutions.

[0006] Firstly, to achieve the above-mentioned purpose, a fire sprinkler head is proposed, which includes: a connecting end, which is installed at the end of a fire pipe; an inner tube, which is connected to the connecting end, supported by the connecting end and sleeved inside the fire pipe, and the side wall of the inner tube is provided with a detection through hole; and, arranged between the connecting end and the inner tube: a sealing disk, which abuts against the end of the fire pipe to prevent the fire sprinkler water from spraying out; a fire glass ball, which is connected between the bottom of the sealing disk and the connecting end, and the fire glass ball supports and limits the sealing disk and the end of the fire pipe to abut tightly; a support rod, which is connected between the top of the sealing disk and the bottom of the inner tube, supports the inner tube and maintains the inner tube at an appropriate height position at the end of the fire pipe; a detection arm, which extends upward from the sealing disk and fits on the outer wall of the fire pipe, the detection arm is marked with a perforation position matching the inner tube detection through hole, and is installed with a sensor detection device.

[0007] Optionally, a fire sprinkler head as described in any of the above, wherein the structural strength of the support rod is weakened at the connection positions at its upper and lower ends; during the installation process, the support rod supports the inner tube from the bottom and maintains the detection hole of the inner tube at the same height as the perforation position on the detection arm; the detection hole is arranged opposite to the detection arm, so that external equipment can penetrate the fire pipe to perform a smoke detection test; after the detection test is completed, the inner tube is restricted by the detection hole and the connecting end is rotated at the same time, so that the sealing disk rotates relative to the support rod, thereby destroying the weakened connection structure at the upper and lower ends of the support rod, and causing the inner tube to fall to the height of the sealing disk.

[0008] Optionally, in any of the fire sprinkler heads described above, the support height of the support rod is not less than the longitudinal inner diameter length of the detection through hole.

[0009] Optionally, in a fire sprinkler head as described above, the upper and lower ends of the fire glass ball are connected to the inner wall of the end by a sealing disk connected by high-elastic UV glue.

[0010] Optionally, a fire sprinkler head as described in any of the above, wherein the connecting end is threadedly connected to the fire pipe end; during the smoke detection test, the threaded connection between the connecting end and the fire pipe end is relatively loose; after the detection test is completed, the threaded connection between the connecting end and the fire pipe end is completely tightened; the fire glass ball and the support rod are both provided with a deflection angle to cooperate with the tightening direction of the connecting end thread.

[0011] Optionally, in any of the fire sprinkler heads described above, the sensing detection device includes: a temperature sensor, a humidity sensor, a smoke sensor, an airflow sensor, a toxic gas sensor, and an oxygen concentration sensor.

[0012] At the same time, in order to achieve the above-mentioned purpose, the present application also provides a detection and early warning system based on fire sprinkler heads, which includes a fire sprinkler head as described in any of the above and a response system communicated with the fire sprinkler head, wherein the response system is provided with a pre-trained: building characteristic model and detection model; the building characteristic model receives the location characteristics of the fire sprinkler head sensor detection device that currently detects an abnormality, and obtains the fire transmission feature matrix corresponding to the location through the weights and parameters obtained in pre-training; the detection model fuses the fire transmission matrix with the sensor features uploaded by each fire sprinkler head whose current detection value changes, and inputs the fused feature matrix into the pre-trained multi-layer attention module and multi-layer perception module for deep training, and finally outputs the fire risk corresponding to each area in the building through the embedding module, and warns the area where the fire risk exceeds the preset indicator.

[0013] Optionally, a detection and early warning system based on fire sprinkler heads as described in any of the above, wherein the building characteristic model is obtained through smoke detection test training; in the smoke detection test, high-temperature smoke simulating a fire is first fed to each fire sprinkler head through a fire pipe in turn, and at the same time, various sensor data of sensor detection devices arranged on the upper and lower floors and the surrounding environment of the fire sprinkler head are tracked, and response functions and weight parameters that reflect the fire situation at the location of each fire sprinkler head and the spread of the fire to the corresponding sensor detection devices on the upper and lower floors and the surrounding environment are trained in the building characteristic model.

[0014] Optionally, a detection and early warning system based on fire sprinkler heads as described in any of the above, wherein, in the detection model, a multi-scale equidistant convolutional network is used to fuse the fire transfer matrix output by the building characteristic model with the sensor features uploaded by each fire sprinkler head whose current detection value has changed; multiple branches using different convolution kernels in the multi-scale equidistant convolutional network are used to model the pattern information contained in the time series of the fire transfer matrix; for each branch, a local module based on downsampling convolution is used to extract local features of the sequence, and on this basis, a global module based on equidistant convolution is used to model the global correlation; finally, a merging operation is used to fuse the information of different patterns from multiple branches, and superimpose it on the sensor features uploaded by each fire sprinkler head whose current detection value has changed.

[0015] Optionally, in a detection and early warning system based on fire sprinkler heads as described above, when the fire transfer matrix obtained by fusing information of different modes from multiple branches is superimposed on the sensor features uploaded by each fire sprinkler head where the current detection value has changed, the Conv2d operator is used to perform a linear fusion operation.

[0016] Beneficial Effects

[0017] The fire sprinkler head provided by the present application and the detection and early warning system based thereon are provided with an inner tube inside the fire pipe, which is supported by a support rod, a sealing plate, and a fire glass ball in sequence, and the inner tube is set at an appropriate height position from the end of the fire pipe by using a connecting section head. The present application uses a sealing plate to set a connecting arm that fits the outer periphery of the fire pipe, and uses the connecting arm to mark the position of the detection through hole of the inner tube on the fire pipe and install the external sensor detection device. The fire sprinkler head of the present application, on the one hand, can use the detection through hole of the inner tube to conduct a smoke detection test to determine the installation effect of each sprinkler head, and another invention can continue to use the corresponding sensor detection device to realize dynamic monitoring of each installation point inside the building after the smoke detection test is completed. The present application can effectively use the detection model to obtain the fire risk of each point in the building area according to the installation point, sensor data and internal characteristics of the building obtained through the previous effectiveness test of each sensor detection device by transmitting the sensor data detected by the sensor detection device in real time to the detection model obtained by pre-training, thereby effectively realizing early warning.

[0018] Other features and advantages of the present application will be set forth in the following description, and in part will become apparent from the description, or may be understood by practicing the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings are used to provide a further understanding of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings:

[0020] Figure 1 It is a schematic diagram of the working mode of the fire sprinkler head of the present application;

[0021] Figure 2 is a schematic diagram of the detection model of the present application;

[0022] In the figure, 1 represents a fire-fighting pipe; 2 represents a connecting end; 3 represents an inner pipe; 30 represents a detection through hole; 21 represents a sealing disk; 20 represents a fire-fighting glass ball; 31 represents a supporting rod; 4 represents a detection arm, and 40 represents a perforation position mark. DETAILED DESCRIPTION

[0023] In order to make the purpose and technical solution of the embodiment of the present application clearer, the technical solution of the embodiment of the present application will be clearly and completely described in conjunction with the drawings of the embodiment of the present application. Obviously, the described embodiment is a part of the embodiment of the present application, not all of the embodiments. Based on the described embodiment of the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0024] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art to which this application belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless defined as herein.

[0025] The meaning of "and / or" described in this application means that the situations where each exists alone or both exist at the same time are included.

[0026] The meaning of "inside" and "outside" in this application refers to the direction pointing to the inside of the fire protection pipe as inside, and the opposite direction as outside; it is not a specific limitation on the device mechanism of this application.

[0027] The term “connection” as used in this application may mean a direct connection between components or an indirect connection between components via other components.

[0028] The meaning of "up and down" mentioned in this application refers to that when the user is facing the fire sprinkler head, the direction from the end of the inner tube to the connecting end provided at the bottom of the fire glass ball is down, and vice versa, and it is not a specific limitation on the device mechanism of this application.

[0029] Figure 1 A fire sprinkler head according to the present application comprises:

[0030] A connecting end 2, which is installed at the end of the fire protection pipe 1;

[0031] An inner tube 3, which is connected to the connecting end 2, supported by the connecting end and sleeved inside the fire protection pipe, and a detection through hole 30 is provided on the side wall of the inner tube;

[0032] And, arranged between the connecting end 2 and the inner tube 3:

[0033] A sealing disc 21 abuts against the end of the fire-fighting pipe to prevent the fire-fighting sprinkler water from spraying out;

[0034] A firefighting glass ball 20 is connected between the bottom of the sealing disk 21 and the connecting end 2, and the firefighting glass ball 20 supports and limits the sealing disk 21 to be in close contact with the end of the firefighting pipe;

[0035] A support rod 31, which is connected between the top of the sealing disk 21 and the bottom of the inner tube 3, supports the inner tube 3 and maintains the inner tube 3 at an appropriate height position at the end of the fire protection pipeline;

[0036] The detection arm 4 extends upward from the sealing disk 21 and fits on the outer wall of the fire protection pipe 1. The detection arm 4 is marked with a perforation position that matches the detection through hole 30 of the inner pipe 3 and is installed with a sensor detection device.

[0037] When installing the fire sprinkler head in this application, first follow Figure 1 In the left-side method, the inner tube 3 is inserted into the fire-fighting pipe 1, and the connecting end 2 is installed at the end of the fire-fighting pipe 1. At this time, the perforation position mark 40 on the detection arm just corresponds to the detection through hole 30 on the inner tube, and an external drilling device can be used to drill a hole on the inner wall of the fire-fighting pipe at this position to form a smoke outlet for smoke detection test.

[0038] In the smoke detection test, each fire-fighting pipe 1 is first arranged in sequence to feed high-temperature smoke simulating a fire to each fire sprinkler head arranged in the building, so as to detect the airflow at the air outlet of the bypass detection pipe through the air outlet formed by the bypass detection pipe on the side wall of the fire-fighting pipe. The detection can be directly realized by the thermal gas flow meter integrated in the sensor detection device. The two thermal resistors used to detect the airflow temperature in the thermal gas flow meter are very small in size, so the inner diameter of the bypass detection pipe can be set to a size that can only accommodate two thermal resistors. Compared with the method that requires the installation of a bulky wind wheel on the bypass pipe, this smoke detection method can greatly reduce the volume of the lateral extension of the airflow detection fire sprinkler head, which is convenient for installation and detection.

[0039] During the smoke detection test, the above-mentioned sprinkler heads and the response system connected to them also track various sensor data of the sensor detection devices arranged on the upper and lower floors and the surrounding environment of the fire sprinkler head. This application pre-integrates the building characteristic model and the detection model in the response system. During the smoke detection test, the building characteristic model can be trained by the change time and degree of change of the sensor data of the adjacent fire-fighting equipment after the sprinkler heads at each point spray, and the response function and weight parameters that can reflect the fire situation at the location of each fire sprinkler head and its spread to the corresponding sensor detection devices on the upper and lower floors and the surrounding environment are constructed in the building characteristic model.

[0040] After the test is completed, a limit rod can be inserted into the test hole 30 of each sprinkler head one by one to limit the circumferential rotation of the inner tube 3, and the connection end 2 can be rotated with the limit rod inserted, so that the connection end 2 drives the sealing disk 21 to rotate relative to the support rod 31, thereby destroying the weakened connection structure at the upper and lower ends of the support rod 31 through the rotation torque, causing the support rod 31 to fail and collapse. After the limit rod in the test hole 30 is removed, the inner tube 3 can fall to the height of the sealing disk 21 by its own gravity, thereby achieving Figure 1The state shown in the middle. In this state, the detection through hole 30 is staggered from the detection channel opened on the side wall of the fire pipe, which can ensure that the fire pipe can remain sealed after the fire sprinkler water supply returns to normal.

[0041] In the present application, the connection end 2 may be provided with a grid having a gap larger than the size of the sealing disk 21. Figure 1 As shown on the right, when the ambient temperature of the sprinkler head of the present application rises and triggers the fire glass ball 20 to break, the broken fire glass ball 20 fragments, the sealing plate 21, and the support rod 31 with a failed connection on the sealing plate can be flushed away from the sprinkler head through the grille by the water flow in the fire pipe. At this time, the inner tube further drops to the grille position. On the one hand, the inner tube provides a water source for the sprinkler head splash plate connected to the bottom of the grille through its pipeline structure, and spreads the water flow in the pipeline in four directions to spray the fire source in the fire protection area; another invention can also provide the water flow splashing out of the pipe to the grille area through the detection through hole 30 on its side wall, thereby adjusting the spraying area of ​​the water flow and spraying a wider range of fire sources.

[0042] In the present application, detection holes 30 can be respectively arranged in each direction of the inner tube, but only one detection hole is kept in the middle of one detection arm. In this way, the detection holes in each direction can provide water sources to the grilles in each direction, so as to achieve the spraying effect on the areas in each direction at different distances from the sprinkler head.

[0043] In order to avoid the fire glass ball 20 from breaking during the installation process and to avoid the support of the inner tube by the support rod 31 cannot be withdrawn after smoke detection, the present application also preferably sets the fire glass ball 20 as two arranged along the circumference of the connecting end 2 at a certain deflection angle in accordance with the tightening direction of the thread of the connecting end 2, and respectively connects the upper and lower ends of the fire glass ball 20 to the inner wall of the connecting end 2 through high-elastic UV glue. The two fire glass balls 20 can share half of the stress when the rotating connecting end drives the sealing disk to rotate. After the high-elastic UV glue is cured by the lamp, a buffer structure is accumulated at the end of the fire glass ball 20. On the one hand, the fire glass ball 20 is stably maintained between the sealing disk and the connecting end, and on the other hand, a certain stress buffer is provided for the fire glass ball 20 when the rotating connecting end drives the sealing disk to rotate, thereby providing protection for the glass structure.

[0044] The lower end of the fire glass ball 20 is respectively arranged at the outer side of the water supply pipe interface connected to the sprinkler water splash plate at the bottom of the connection end 2, close to the bottom center of the connection end 2. When the fire glass ball 20 is broken, it will fall down along the circumference of the connection end 2 due to the rotation angle set by it. The bottom of the fire glass ball 20 is wrapped with high elastic UV glue to prevent its broken fragments from falling into the sprinkler water splash plate water supply pipe at the bottom of the connection end 2. The upper end of the fire glass ball 20 diverges outward and is connected to a position close to the outer peripheral edge of the sealing disk to keep the sealing disk and the supporting rod connected above it as stable as possible, and to ensure that the residue when it is broken can be as close to the grid gap of the outer wall of the connection end 2 as possible and be washed away by the water flow in the pipe to ensure the spraying effect.

[0045] A layer of elastic material can be attached to the upper surface of the sealing disk 21, which can be made of silicone or rubber and can be set to be hollow or solid. On the one hand, the elastic material can provide a certain sealing effect by using its thickness to make its upper surface abut against the end of the fire pipe earlier when the connecting end 2 is not fully tightened during smoke detection; on the other hand, it can also provide a certain elastic deformation stroke range when the connecting end 2 is fully tightened after the smoke test, and absorb the stress of the fire glass ball 20, ensuring the sealing of the device and ensuring that the fire glass ball 20 will not fail due to excessive force.

[0046] In order to ensure that the threaded connection between the connecting end 2 and the end of the fire-fighting pipe 1 after the smoke detection test is completed, the upper and lower connection structures of the support rod 31 can be completely invalidated during the process of fully tightening the threaded connection between the connecting end 2 and the end of the fire-fighting pipe 1, the present application also preferably weakens the structural strength of the support rod 31 at the connection positions at the upper and lower ends, and the support height of the support rod 31 is not less than the longitudinal inner diameter length of the detection through hole 30 to ensure that when the inner tube falls on the sealing disk, the detection through hole can be completely offset from the drilling position on the fire-fighting pipe wall. During the smoke detection test, the threaded connection between the connecting end 2 and the end of the fire-fighting pipe 1 is relatively loose, and the weakened structure at both ends of the support rod 31 is less stressed, so the support rod 31 can support the inner tube 3 from the bottom and maintain the detection through hole 30 of the inner tube 3 at the same height as the perforation position on the detection arm 4. After the experiment, when tightening the connecting end 2, due to the weakened structure at both ends of the support rod 31, this position can be easily torn off. At this time, the support rod 31 rotates and falls along the deflection angle of the thread tightening direction of the matching connection end 2 to ensure that the downward travel range of the inner tube is as large as possible to avoid the opening position of the fire pipe. At the same time, since the installation direction of the fire pipe has the same rotation direction, the stress generated when it is restricted by the sealing disk is a pulling force pulling the two ends of the fire glass ball 20. This pulling force can be contracted by the elastic deformation of the high-elastic UV glue to avoid failure caused by stress squeezing the glass ball.

[0047] The sensing detection device on the detection arm of the present application can integrate: temperature sensor, humidity sensor, smoke sensor, airflow sensor, toxic gas sensor, oxygen concentration sensor. It can be powered by the solar photovoltaic module set on the detection arm, connected to the Bluetooth or wireless interaction unit to achieve interaction through wireless signals, and upload the real-time environmental characteristics collected by each sensor and the coding information and / or location information of the sensor to the response system.

[0048] The response system receives the location feature of the fire sprinkler head sensor detection device that currently detects abnormality through the building characteristic model obtained in advance training, and obtains the fire transmission feature matrix corresponding to the location through the weights and parameters obtained in advance training;

[0049] The detection model is then used to fuse the fire transfer matrix with the sensor features uploaded by each fire sprinkler head whose current detection value has changed. The fused feature matrix is ​​input into the multi-layer attention module and multi-layer perception module of the pre-trained activity for deep training. Finally, the fire risk corresponding to each area in the building is output through the embedding module, and an early warning is issued for areas where the fire risk exceeds the preset indicator.

[0050] Specific reference Figure 2 The building characteristic model of the detection and early warning system of the present application is trained and obtained in the following manner:

[0051] In the smoke detection test, high-temperature smoke simulating a fire is first fed to each fire sprinkler head through the fire pipe 1 in turn, and the exhaust volume of the detection hole in the fire sprinkler head currently fed with smoke is used to confirm whether the sprinkler head can operate normally. At the same time, various sensor data of the sensor detection devices arranged in the upper and lower floors, surrounding environment, and adjacent rooms of the fire sprinkler head are tracked. When the detection devices on other sprinkler heads detect changes in environmental sensor parameters caused by smoke propagation, the position characteristics corresponding to the detection devices affected by the environmental changes are recorded, such as the installation floor, room number, and relative position in the room, and the interval time of the change relative to the smoke feeding of the test sprinkler head is marked. The position characteristics of the sprinkler head fed with smoke and the position characteristics of other sensor devices affected by it are encoded into a matrix according to the preset encoding rules, and are input into the building characteristic model. The attention module (attention) in the building characteristic model processes the trigger time information, position information, and relative information of the sensor devices contained in the matrix. For the propagation channel characteristics of other sensor devices in the building, the matrix output by the attention module that reflects the above characteristics is further input into the feedforward layer in the building characteristic model. By comparing the output results of the feedforward layer for the position characteristics of the sprinkler head fed with smoke with the influence of other position sensor devices (the influence is obtained by processing the encoding matrix by the attention module), the parameter matrix of each module in the building characteristic model is trained and adjusted, so as to obtain a parameter matrix that can converge so that the output results of the feedforward layer for the position characteristics of the sprinkler head fed with smoke are close to the influence of other position sensor devices. Thus, the response function and weight parameters of the fire situation at the location of each fire sprinkler head in the building characteristic model that affects the upper and lower floors and the surrounding environment are determined.

[0052] In order to ensure that the attention module in the above-mentioned building characteristic model can simultaneously obtain the trigger time information, position information and propagation channel characteristics of the sensor devices contained in the matrix relative to other sensor devices in the building, the attention module of the convolutional neural network used in this application can be specifically set to include two branches:

[0053] The first branch includes a first deformable convolution layer, a first convolution layer, and a first global pooling layer connected in sequence, and the output of the first deformable convolution layer is set to:

[0054]

[0055] Correspondingly, the second branch includes a second deformable convolutional layer, a second convolutional layer, and a second global pooling layer connected in sequence, and the output of the second deformable convolutional layer is:

[0056]

[0057] Among them, p0 represents each feature point in the input feature matrix; f(p j ) and f(p i ) represent the position propagation channel attributes and position information of the feature point p0 in two spatial dimensions after the deformable convolution operation; R 3 represents the convolution kernel, p n is an enumeration of the positions in the convolution kernel; Δp i and Δp j Represents the offset in two dimensions respectively.

[0058] Then, the outputs of the two branches are concatenated to obtain a concatenated matrix, and the concatenated matrix is ​​transformed; the transformed concatenated matrix is ​​fed to the fully connected layer to perform convolution operations on it in different dimensions, and the attention vector of the input matrix corresponding to the propagation channel dimension and the coordinate dimension is obtained; based on the attention vector, each value in the feature matrix is ​​weighted.

[0059] Therefore, the present application can pool the feature matrix composed of the sensing information on each floor and in each direction and the feature matrix composed of the sensing information in the vertical direction across floors respectively through the above two branches. Through pooling, the propagation channel properties of the long-range mutual dependence and mutual influence between the sensing detection devices are captured along a spatial dimension in the characteristic matrix output by it, and at the same time, the precise triggering corresponding position information is retained along another empty dimension in the characteristic matrix output by it, so that the sensor triggering information of each time dimension and coordinate dimension in the entire building space can be retained, and at the same time, the attention of the building space and the attention of the fire spread channel corresponding to the environmental layout inside the building are integrated, which solves the problem that the existing attention mechanism cannot fully adapt to the respective extraction needs in the space and propagation process due to unified operation, so as to improve the accuracy of the convolutional neural network.

[0060] The position feature of the fire sprinkler head sensor detection device currently detecting an abnormality is received, and the fire transmission feature matrix corresponding to the position is obtained by pre-training weights and parameters. Then, in the detection model, a multi-scale isometric convolution network (MICN) is used to fuse the fire transmission matrix output by the building characteristic model with the sensor features uploaded by each fire sprinkler head whose current detection value changes;

[0061] In the multi-scale equidistant convolutional network, different convolution kernels shown in Table 1 are used to construct multiple branches, and the pattern information contained in the time series of the fire transfer matrix is ​​modeled according to different time steps;

[0062] For each branch, a local module based on downsampling convolution is used to extract local features of the sequence. On this basis, a global module based on equidistant convolution is used to model the global correlation.

[0063] Finally, a merge operation is used to fuse the information of different modes from multiple branches and superimpose them on the sensor features uploaded by each fire sprinkler head whose current detection value has changed.

[0064] Thus, the fused sensor features are input to the multi-layer attention module and perception module of the subsequent stage, and the attention module is used to extract the information of the channel characteristics of each sensor data and the corresponding sensor location in the fused features layer by layer, and the information is processed in parallel with the context sensor data and channel characteristics through the perception module, so as to embed the output matrix of the last layer of perception module, obtain the warning information based on the fire transmission characteristics of each sensor data and the channel where each sensor is located, and obtain the probability prediction value of the fire at the location of each sensor, and then output the warning information for the area when the prediction value reaches or exceeds the preset index, and send the warning information to the on-duty personnel terminal in the area through the wireless network. In a more preferred mode, the abnormal situation targeted by the warning information can also be prompted or directly triggered to ventilate, cool, humidify, or close doors and windows of the smart devices installed in the warning area, so as to minimize the possible fire hazards.

[0065] In addition, in the aforementioned process, the present application can also use the Conv2d operator to perform linear fusion operations through convolution when the fire transfer matrix obtained by fusing information from different modes of multiple branches is superimposed on the sensor features uploaded by each fire sprinkler head whose current detection value has changed. In this step, multiple layers of Conv2d operators are set for convolution, and the Sigmoid activation function is used in its output layer, and the Relu activation function is used in the remaining layers. Conv2D convolution can extract features from its input data, and at the same time, through this operation, the number of features required to be processed by the next convolution layer is reduced, thereby reducing the operation parameters and speeding up the calculation speed. The convolution kernel sizes of Conv1 to Conv8 in the Conv2D convolution layer are all (3,3), and the number of convolution kernels is shown in Table 1: the convolution kernel size of Conv9 is (1,1), and the number of convolution kernels is 1; the step size of each convolution is (1,1), and each convolution layer contains a batch normalization layer and a Relu activation function. In some implementations, the present application may also add two residual modules of the same structure after each convolutional layer of Conv1 to Conv8, and add an attention module after every two residual blocks to deepen the network depth, achieve cross-channel information fusion, and overcome the problems of network gradient disappearance and performance degradation caused by deepening the network depth due to the residual structure.

[0066] Table 1

[0067] Convolutional Layer Convolution kernel size Number of convolution kernels Conv1 3*3 16 Conv2 3*3 32 Conv3 3*3 64 Conv4 3*3 128 Conv5 3*3 128 Conv6 3*3 64 Conv7 3*3 32 Conv8 3*3 16 Conv9 1*1 1

[0068] The above is only an implementation method of the present application, and its description is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present application. It should be pointed out that for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present application, which all belong to the protection scope of the present application.

Claims

1. A fire sprinkler head, characterized in that: include: A connecting end (2) mounted at the end of the fire fighting pipe (1); An inner tube (3) is connected to the connection end (2), supported by the connection end and sleeved inside the fire protection pipe, and a detection through hole (30) is provided on the side wall of the inner tube; And, arranged between the connecting end (2) and the inner tube (3): A sealing disc (21) abuts against the end of the fire-fighting pipe to prevent the fire-fighting sprinkler water from spraying out; A firefighting glass ball (20) connected between the bottom of the sealing plate (21) and the connecting end (2), the firefighting glass ball (20) supporting and limiting the sealing plate (21) and the end of the firefighting pipe to be in close contact with each other; A support rod (31) connected between the top of the sealing disk (21) and the bottom of the inner tube (3) to support the inner tube (3) and maintain the inner tube (3) at an appropriate height position at the end of the fire protection pipeline; The detection arm (4) extends upward from the sealing disk (21) and fits on the outer wall of the fire protection pipe (1). The detection arm (4) is marked with a perforation position that matches the detection through hole (30) of the inner pipe (3) and is equipped with a sensor detection device.

2. The fire sprinkler head according to claim 1, characterized in that: The structural strength of the support rod (31) is weakened at the connection positions at the upper and lower ends thereof; During the installation process, the support rod (31) supports the inner tube (3) from the bottom and maintains the detection through hole (30) of the inner tube (3) at the same height as the perforation position on the detection arm (4); The detection through hole (30) is arranged opposite to the detection arm (4) so ​​as to allow external equipment to penetrate the fire protection pipe (1) to perform a smoke detection test; After the detection test is completed, the inner tube (3) is restricted by the detection through hole (30) and the connecting end (2) is rotated at the same time, so that the sealing disk (21) rotates relative to the support rod (31), thereby destroying the weakened connection structure at the upper and lower ends of the support rod (31), and causing the inner tube (3) to fall to the height of the sealing disk (21).

3. The fire sprinkler head according to claim 2, characterized in that: The supporting height of the supporting rod (31) is not less than the longitudinal inner diameter length of the detection through hole (30).

4. The fire sprinkler head according to claims 1-3, characterized in that: The upper and lower ends of the fire-fighting glass ball (20) are connected to the inner wall of the end head (2) through a sealing disk (21) connected to the end head (2) via a high-elasticity UV adhesive.

5. The fire sprinkler head according to claims 1-4, characterized in that: The connection end (2) is threadedly connected to the end of the fire-fighting pipe (1); During the smoke detection test, the threaded connection between the connection end (2) and the end of the fire-fighting pipe (1) is relatively loose; after the detection test is completed, the threaded connection between the connection end (2) and the end of the fire-fighting pipe (1) is completely tightened; The firefighting glass ball (20) and the support rod (31) are both provided with a deflection angle that matches the thread tightening direction of the connecting end (2).

6. The fire sprinkler head according to claim 5, characterized in that: The sensing detection device includes: a temperature sensor, a humidity sensor, a smoke sensor, an airflow sensor, a toxic gas sensor, and an oxygen concentration sensor.

7. A detection and early warning system based on fire sprinkler heads, characterized in that: It comprises a fire sprinkler head as claimed in any one of claims 1 to 6 and a response system connected to the fire sprinkler head in communication therewith, wherein the response system is provided with: a building characteristic model and a detection model obtained by pre-training; The building characteristic model receives the location feature of the fire sprinkler head sensor detection device that currently detects an abnormality, and obtains the fire transmission feature matrix corresponding to the location through weights and parameters obtained by pre-training; The detection model combines the fire transfer matrix with the sensor features uploaded by each fire sprinkler head whose current detection value changes, and inputs the fused feature matrix into the multi-layer attention module and multi-layer perception module of the pre-trained activity for deep training. Finally, the fire risk corresponding to each area in the building is output through the embedding module, and an early warning is issued for areas where the fire risk exceeds the preset indicator.

8. The fire sprinkler head-based detection and early warning system according to claim 7, characterized in that: The building characteristic model is obtained through smoke detection test training; In the smoke detection test, high-temperature smoke simulating a fire is first fed to each fire sprinkler head through the fire pipe (1) in sequence, and at the same time, various sensor data of the sensor detection devices arranged on the upper and lower floors of the fire sprinkler head and in the surrounding environment are tracked, and the response function and weight parameters that reflect the fire situation at the location of each fire sprinkler head and its upper and lower floors and surrounding environment are trained in the building characteristic model to obtain the corresponding sensor detection devices.

9. The fire sprinkler head-based detection and early warning system as described in claims 7-8, characterized in that: In the detection model, a multi-scale equidistant convolutional network is used to fuse the fire transfer matrix output by the building characteristic model with the sensor features uploaded by each fire sprinkler head whose current detection value changes; Multiple branches with different convolution kernels are used in the multi-scale equidistant convolutional network to model the pattern information contained in the time series of the fire transfer matrix. For each branch, a local module based on downsampling convolution is used to extract local features of the sequence. On this basis, a global module based on equidistant convolution is used to model the global correlation. Finally, a merge operation is used to fuse the information of different modes from multiple branches and superimpose them on the sensor features uploaded by each fire sprinkler head whose current detection value has changed.

10. The fire sprinkler head-based detection and early warning system according to claims 7-9, characterized in that: When the fire transfer matrix obtained by fusing information from different modes of multiple branches is superimposed on the sensor features uploaded by each fire sprinkler head whose current detection value changes, the Conv2d operator is used to perform linear fusion operations.

Images